A MINLP Model for Crude Oil Selection and Refining

CATEGORY: CRUDE OIL SCHEDULING
Proceedings of GOW - Proceedings Of The Global Optimizationworkshop 2012, pp. 57 – 60
A MINLP Model for Crude Oil Selection and Refining
Ormeu Coelho ,1;2  Laura Bahiense, 2 and Virgílio J. M. Ferreira Filho 2;3
1 CEFET-RJ / Celso Suckow da Fonseca, Department of Production Engineering, R. General Canabarro, 229, CEP 20271-110,
Rio de Janeiro-RJ, Brazil, ormeucoelho@gmail.com
2 COPPE-UFRJ, Production Engineering Program, Rio de Janeiro-RJ, Brazil, laura.bahiense@gmail.com
3 UFRJ, Polithenic School, Department of Production Engineering, Rio de Janeiro-RJ, Brazil, virgilio@ufrj.br
Abstract
Authors solve a real-based refining planning problem by MINLP techniques. The developed model aims to maximize the refiner profity by integrating decisions on operation of processing units and crude oil selection. An illustrative instance was solved by the global optimizer BARON 7.5.3 and by the (heuristic) AOA, both avaible in AIMMS 3.11.
Introduction
The model presented in this work is an extension of the multiperiod formulation proposed by [1]. In each period of the planning horizon there are deterministic demands for oil derivatives (expressed as minimum and maximal volumes to be sold), and available volumes of crude oils and other inputs. For each crude oil two quantities may be available. A quantity that refinery must receive from supply source but not necessarily consume. And additional quantities that could be ordered if necessary, but within certain limits. Stocks of all streams (crude oils/inputs, intermediate and final products) could be carried from one period to the following. The objective function aims to maximize the profit defined by revenues obtained by derivative sales minus costs incurred with raw material consumption and storage in tanks. For doing so, it requires the computation of flows among units and stock levels in each tank, its physicochemical properties, and operational variables on conversion units. Some decisions about flows have a discrete behaviour since it is usually required that they be bigger than a predefined minimum values. In this sense, in each planning period, one should determine which campaigns will be allocated to processing units and what kinds of additional crude oils (inputs) should be ordered. Preliminare results with this formulation were published in [2]. Given the large number equations defined by sums of bilinear and tri-linear terms, the resulting non-convex MINLP is highly nonlinear. Even having lot less integer variables than continuous ones, the generated programs are hard to solve to optimality.  
Free Full Text Source: http://www.hpca.ual.es/~leo/gow/2012-XI-GOW.pdf#page=65